Understanding Google AI Mode: How It Works, Access and SEO Impact

When I first wrote this article in October 2025, AI Mode was a few days old in Germany. A year later it is no longer a trial balloon.

Search interest shows how far this has shifted. In August 2025 roughly 1,900 people per month searched for „ai mode” in Germany. In July 2026 it was 74,000, almost forty times as many (DataForSEO, retrieved 13 August 2026, search volumes are estimates). That is not a curve, that is a wall. What people were actually after was visible on the search results page that same day: the leading positions for „ai mode” were almost entirely explainers and glossary entries, from Google’s own product page to Wikipedia. People are looking for a definition.

That is what this article delivers. It explains AI Mode from how it works to how you access it in Germany and how it differs from AI Overviews. After that it gets uncomfortable. The numbers on click behaviour are clear and they are better documented than they were a year ago. The practical part comes at the end.

One note up front, because there is a lot of half-knowledge around AI search: every statement about rollout, feature scope and requirements in this article is taken from Google’s own documentation. Where I cite studies, I name the sample size and the period. Both have become more important than they sound.

Google AI Mode is called KI-Modus in German and is an interactive chat inside Google Search. Instead of a list of ten blue links you get one coherent AI answer and below it the links to the sources it was assembled from.

The decisive word is interactive. You can ask again and have an answer sharpened or shift the focus or add a constraint, without restating your question from scratch every time. Classic search forgets your previous query, AI Mode does not.

How to open AI Mode

Google enabled AI Mode for Germany, Austria and Switzerland on 8 October 2025, together with over 40 further countries and 36 new languages. There are three ways in:

• via the „AI Mode” tab on the search results page, next to „All”, „Images” and „Videos”
• directly via google.com/ai
• via the „AI Mode” button in the Google app

Query fan-out: one question becomes many

The most striking technical difference to classic search is called query fan-out. Instead of running a query against the index once, the AI breaks it into several sub-questions and answers them in parallel and builds one coherent answer from the partial results.

An example. The input „best noise-cancelling headphones for train journeys under 300 euros” does not trigger one search, but several. The AI then researches separately for current test reports and for noise suppression on trains and for battery life within the price limit. What you read as a single answer is the consolidation of all these partial searches.

For SEO this has an uncomfortable consequence. Your page no longer has to be the best for one query, it has to be the best source for one of the sub-questions. How Google processes content for this in the first place is something I wrote up in detail in my article on how AI Overviews work. The processing path is the same.

Multimodal input

AI Mode processes several types of input at the same time. You can type text, speak or add an image and combine all of it. Camera search via Google Lens is a fixed part of it.

This is more than a convenience question. When a user uploads a photo of a component and asks which spare part fits, that query was never formulated as a keyword. You cannot optimise for that in the classic sense and you can only make sure that your own content delivers the matching answer cleanly and unambiguously.

The two features get confused constantly and that is simply because both appear as AI text above the results. The difference lies in who triggers them and what happens afterwards.

AI Overviews (officially „AI Overview”) are automatically generated text blocks that answer a query directly above the organic results. You do not have to do anything. They appear on their own when Google’s systems consider them the most helpful answer to the question asked. In Germany they have been running since 26 March 2025.

AI Mode on the other hand only happens when you open it. In return you get a dialogue instead of a summary.

In practice this means AI Overviews reach you on every normal search, AI Mode only on those where someone deliberately opens it. In terms of reach the Overviews are therefore the bigger topic, AI Mode the faster growing one.

• Data access: AI Mode builds on Google’s index. Google explicitly ties visibility in AI Overviews and AI Mode to indexing, so the currency of the answers hangs on the same pipeline as classic search. How ChatGPT and Perplexity assemble their answers in detail is described differently by each provider and in changing stages of development. That is exactly where comparison tables go stale fastest.
• Location: AI Mode is a tab in a product that is used daily. The other two are standalone applications you actively switch to. This difference decides reach, not answer quality.
• Purpose: My reading after a year of use: AI Mode is built as a research path inside search, not as a general-purpose assistant. ChatGPT covers a considerably broader range of tasks. That is an assessment from practice and not a vendor statement.

This is where it gets concrete and where a closer look pays off. On the topic of zero-click there are numbers in circulation that are partly two years old and are still passed on in talks and articles as the current state.

The general zero-click trend

The most recent broad study comes from SparkToro, based on the Similarweb panel. For January to April 2026, 68.01 percent of all Google searches in the US ended without a click. For comparison, the same analysis puts 2024 at 60.45 percent. That is an increase of 7.56 percentage points in two years, according to SparkToro the fastest acceleration of the phenomenon in the past decade. To the analysis

Click behaviour inside AI Mode specifically

For AI Mode there is now a large analysis. Eric Van Buskirk of Clickstream Solutions analysed anonymised clickstream data from Surfer SEO, roughly 846,000 US search sessions from February and March 2026. The result: in 88 percent of AI Mode tasks the shortlist proposed by the AI was accepted as it stood, in 74 percent the top-listed entry was the one chosen, and in 64 percent no link was clicked at all. To the summary at Search Engine Land

This number replaced an earlier one that circulated through the industry for a long time and rested on 250 sessions. 846,000 against 250 is not a detail, it is the difference between an observation and a measurement.

And in the German-speaking market?

For the German-speaking market the data situation is thinner. An analysis by Wordsmattr.io covering 125 SEO-optimised landing pages in Germany, Austria and Switzerland arrives at an average of 17.8 percent fewer clicks since the AI Overviews launch on 26 March 2025, with impressions almost unchanged. To the study

125 pages are a narrow basis. The direction matches the larger US studies, the size of the effect should not be derived from it. What you can do instead: in Search Console, put the development of impressions and clicks for your own top pages side by side. If impressions stay stable while clicks fall, you are seeing the effect in your own data instead of someone else’s. How to go about it is described in my article on zero-click search.

Until this update, this section said that structured data was the decisive lever for being cited in AI Overviews. I have been running into that assumption in guides and conference talks for months. I deleted the sentence after putting Google’s own documentation next to it. It says the opposite, in as many words.

The documentation on AI features states: „You don’t need to create new machine readable files, AI text files, or markup to appear in these features. There’s also no special schema.org structured data that you need to add.” And further: „There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.” Google Search Central

What is actually required is stated there as well: „To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements.” That is the entire entry ticket.

From this follows a shorter and more honest list:

• Secure the technical basis. Indexability and snippet eligibility are not optional extras, they are the precondition. Anyone blocking via nosnippet or through a faulty robots.txt removes their own foundation.
• Deliver content that cannot be summarised away. An AI can aggregate what exists. Your own studies, your own measurements and your own analyses do not exist beforehand. That is the one advantage an aggregator does not catch up with.
• Answer sub-questions, not keywords. When the query fan-out breaks a request apart, the text that wins is the one answering one of those sub-questions cleanly and completely. Clear subheadings and answered questions help more than keyword density.
• Keep maintaining structured data, but for the right reason. It is not a condition for being cited in AI answers. For rich results in classic search it remains relevant. The details are in my guide to structured data and AI Overviews.
• Make experience and authorship visible. Traceable author profiles, named sources and disclosed methodology are good practice. Google describes E-E-A-T in the Quality Rater Guidelines as what human raters judge by. It is not a named ranking signal of the AI systems. More on this in my E-E-A-T guide.
• Measure the effect in your own data. Rank trackers fall short here because they measure a position that plays no role in AI Mode. Search Console on the other hand shows what actually happens to impressions and clicks on your pages.

Between August 2025 and July 2026 demand for „ai mode” rose from 1,900 to 74,000 monthly searches. That alone does not make the feature important, but it does show that it has left the niche.

What changes for daily work is more manageable than the tone of some debates suggests. Clicks are falling, that is measured. The entry conditions have stayed the same and anyone who has worked solidly in recent years therefore does not have to overturn anything: indexable, snippet-eligible and editorially distinct was already the job before.

The most honest advice I can give: on this topic, do not rely on quoted percentages, including the ones in this article. Look into your own Search Console and compare impressions with clicks. That is where you see what is actually happening to you.